David Brown wrote:
Most of his complaints about performance concern read performance. The 'stat' call on Linux is about 1,000 times faster than something equivalent under Windows.
I vaguely recall something odd about the way Linux does stat, but I can't remember it. It also comes up in the context of why Linux sucks at NFS because suddenly it has to obey some obscure semantics that it normally blows off.
I wasn't aware that Linus liked thrashed filesystems? ext2/3 have had fairly conservative write caching for quite some time.
Yeah, he does. He bloviates about softupdates (which tries to straddle the sync/async tradeoff slightly differently) in *BSD every now and then. Linux tends to mount everything async unless you go out of your way. Unsurprisingly, when you force things in Linux to mount sync, or allow other systems to mount async, the performance numbers generally move pretty close to one another.
Unsurprising, really. A good filesystem should be limited by the characteristics of the disk given how much CPU and memory can be thrown at the problem.
I can't speak to Windows, though. -a -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
